Repository that contains some testing utilities for the ergocub project.
This repo contains a gazebo
model, starting from iCub3, and some scripts that can easily modify it and test it in diverse scenarios. The goal of this repo is to gain knowledge about which is the best configuration for the ergoCub in terms of limb length, mass, etc.
- Nicolo Genesio (@Nicogene)
gazebo
robotology-superbuild
v2022.02.0 withdynamics profile
enabled andROBOTOLOGY_USES_GAZEBO=ON
.- For running the walking and balancing experiment the superbuild dynamics-full-deps profile and the flag
ROBOTOLOGY_USES_MATLAB
have to be enabled, alongside a matlab installation. urchin
dataclasses
- Clone this repo and create a
build
directory on it:
git clone https://github.com/icub-tech-iit/ergocub-gazebo-simulations.git
cd ergocub-gazebo-simulations
mkdir build
cd build
- Then use
ccmake
to specify the directory where you installedrobotology-superbuild
:
ccmake ..
CMAKE_INSTALL_PREFIX <directory-where-you-downloaded-robotology-superbuild>/build/install
- You can now proceed to build:
make install -j4
You should now see three new models when you open gazebo
from the terminal:
Finally, add export YARP_ROBOT_NAME=stickBot
in your .bashrc
.
If you want to use ergocub-gazebo-simulations
and you want to install gazebo-yarp-plugins
and other dependencies from conda binaries, you can first of all create a new environment (you can skip this step if you already have an environment you want to use):
mamba create -n ergocubsim
Then, you can activate the environment and install the dependencies:
mamba activate ergocubsim
mamba install -c conda-forge -c robotology robotology-distro yarp icub-models gazebo-yarp-plugins gazebo whole-body-controllers walking-controllers cmake ninja whole-body-estimators
Clone this repo and create a build
directory on it:
git clone https://github.com/icub-tech-iit/ergocub-gazebo-simulations.git
cd ergocub-gazebo-simulations
mkdir build
cd build
After that, you can install the repo with these commands on Windows:
cmake -GNinja -DCMAKE_INSTALL_PREFIX=%CONDA_PREFIX%\Library ..
ninja install
and with this command on Linux/macOS:
cmake -GNinja -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX ..
ninja install
You should now see three new models when you open gazebo
from the terminal:
Finally, set the YARP_ROBOT_NAME
in your environment to stickBot
, via the command:
conda env config vars set YARP_ROBOT_NAME=stickBot
After this, make sure that you re-activate your conda environment before using the simulator and the software that use the simulator.
cd scripts
./modify_robot.py # [-c {PATH}] [-r]
The modify_robot
script changes the dimensions of the stickbot according to a configuration file. It also performs the installation step, so gazebo gets updated with the newest robot right away.
Flags:
-c
/--config {PATH}
Path of the configuration file, default is theconf.ini
file insidescripts
.-r
/--reset
Setting this flag returns the robot to its latest version from git (by usinggit-checkout
). This flag is prioritized over the configuration file.
The conf.ini
file is a simple configuration file specifying the elements to modify:
[left_arm]
dimension = 0.65
radius = 0.45
density_scale = 1.5
[r_lower_leg]
dimension = 1.5
The upper section (the one inside []
) specifies the element to modify, options include torso, right_arm, left_arm, right_leg, left_leg, arms, legs
, all
, or the name of an individual link (Caution: if a link and its respective limb are both specified, the changes could accumulate!). Sections with other names will be ignored.
For any element you can specify dimension
, mass
, density
, and radius
, which will set the specific property for all the valid links. If you want relative scaling you can instead add the _scale
suffix and the factor to multiply to the original value. If a config section includes both an absolute and a scaling modifier the absolute one takes precedence. Mass modifications take precendence over density.
This sandbox uses the worldInterface
gazebo yarp plugin. To run it:
cd experiments/hold_box
./hold_box.sh
This experiment creates a StickBot in gazebo
with its hands open. Then, a box spawns on top and the robot holds it:
ergocub_first_test.mp4
Instead of closing every window that opens after the test, you can call the cleanup script:
./cleanup.sh
To change the properties of the box being spawned you can call the modify_box
script:
./modify_box.py # [-p {X Y Z}] [-d {D W H}] [-m {MASS}] [-r]
After running the test, a new directory called telemetry_data
will appear, use the plotTelemetry.m
script in MATLAB to plot the results for the specific joints you need:
Flags:
-p
/--position
The spawning position of the box, inputted as three numbers representing X, Y and Z coordinates-d
/--dimensions
The dimensions of the box, inputted as three numbers representing its depth, width and height-m
/--mass
The mass of the box as a float-r
/--reset
Setting this flag returns the box to its latest version from git (by usinggit-checkout
). This flag is prioritized over all other flags.
This experiment attaches weights to the stickBot and makes it perform gym-like motions, in order to isolate (to the stickBot's best abilities) the effort to single joints. This sandbox uses the worldInterface
gazebo yarp plugin, then before running the script you must append to GAZEBO_MODEL_PATH
the /my/workspace/robotology-superbuild/src/GazeboYARPPlugins/tutorial/model
path. Then:
cd experiments/gym
./gym_session.sh # {curls lateral_raises front_raises}
The three possible experiments are the following:
After running the test, a new directory called telemetry_data
will appear, use the plotTelemetry.m
script in MATLAB to plot the results for the specific joints you need:
These experiments implement walking and balancing, to stress the leg joints.
For running the walking experiment, just run the go_for_a_walk.sh
script.
cd experiments/walking
./go_for_a_walk.sh
Gazebo will spawn, and the robot will walk like so:
stickbot-walk.mp4
For this to work, WalkingModule
needs to be installed in the system.
It is important to mention that the behavior can be unreliable, and the robot may some times fall. For this reason, the script runs Gazebo with a GUI (unlike the other experiments). This way one can run the experiment several times until a successful walk is achieved. Once the robot walks properly, you will have the data available.
The balancing experiment is ran using whole-body-controllers. Particularly, the floating-base-balancing-torque-control demo. You should launch Gazebo, spawn a stickbot and then run the simulink model. From there, you can use the plotTelemetry.m
matlab script to visualize the data in the joints of interest.
stickbot-balancing.mp4
For bugs and feature requests feel free to submit an issue and we will look into it.